Dimension reduction analysis in image-based species classification

Automated insect species identification and classification is of great demand in agriculture, environment and entomology areas these days. In the paper, an image-based species classification method is proposed, which includes a snake-based segmentation method and global feature extraction. For global feature data, three dimension reduction analysis methods are proposed to get more effective classification feature data, which are regression and correlation analysis, principal component analysis and multivariate analysis of variance. Species samples are selected from nine different species, which are the most harmful insect species in the agriculture orchard. Five classifiers, minimum least square linear classifier, normal density based linear classifier, K nearest neighbor classifier, Parzen density based classifier and nearest mean classifier are compared for feature data from different dimension reduction analysis methods, and the best performance is given by normal density based linear classifier. The results show that different dimension reduction methods give different classification performance according to different classifiers. Compared to the classification results before and after taking dimension reduction analysis, suitable dimension reduction method can be used to keep both efficiency and effect of classification.